Professional Documents
Culture Documents
Anthony Aguirre
School of Natural Sciences, Institute for Advanced Study Princeton, New Jersey 08540, USA∗
Steven Gratton
Joseph Henry Laboratories, Princeton University, Princeton, New Jersey 08544, USA†
(Dated: February 7, 2008)
We develop our recent suggestion that inflation may be made past eternal, so that there is no initial
cosmological singularity or “beginning of time”. Inflation with multiple vacua generically approaches
a steady-state statistical distribution of regions at these vacua, and our model follows directly from
making this distribution hold at all times. We find that this corresponds (at the semi-classical level)
to particularly simple cosmological boundary conditions on an infinite null surface near which the
spacetime looks de Sitter. The model admits an interesting arrow of time that is well-defined and
arXiv:gr-qc/0301042v2 28 Mar 2003
consistent for all physical observers that can communicate, even while the statistical description
of the entire universe admits a symmetry that includes time-reversal. Our model suggests, but
does not require, the identification of antipodal points on the manifold. The resulting “elliptic” de
Sitter spacetime has interesting classical and quantum properties. The proposal may be generalized
to other inflationary potentials, or to boundary conditions that give semi-eternal but non-singular
cosmologies.
A semi-eternally inflating cosmology naturally arises is the physical radius at t of a bubble nucleated at t0 .
from a generic double-well potential [9, 10]. Consider Third, we may calculate, for a given point in the in-
some large comoving region in which, at an initial time flating region at time, the number per unit time N (r, t)
t = t0 , the energy density is dominated by the inflaton of incoming bubbles of physical radius r. This is
φ, with φ ≃ φf and ∂µ φ ≃ 0. Assume the spacetime to
locally resemble de Sitter (“dS”) space, with (for conve- 4πλr2
N (r, t) = (5)
nience) nearly-flat spatial sections, i.e. with a metric [11] (1 + Hr)4
3
spreads, and hx̂i increases, eventually approaching the A. The singularity theorems
classical trajectory for hx̂i ≫ xf (we could say roughly
that the particle has “decayed” when hx̂i changes sig- Having examined the AOT, we may turn to the sin-
nificantly from xf .) Note also that the same spreading gularity theorems, for the AOT proves crucial in their
would occur at t < 0 so that this purely quantum de- analysis. Both the older theorems [3] and the newer theo-
scription would be T-symmetric. Now, a semi-classical rem [4] assert that if certain conditions are satisfied every-
description of the system would describe the system clas- where within a spacetime, then not all past non-spacelike
sically at both early and late times, but with a quantum- geodesics have infinite proper time or affine parameter.
mechanical transition connecting the classical trajecto- This indicates that the spacetime is either extendible or
ries at some given time. This transition time is random, contains singularities.
and in an ensemble of such systems (as is required for The older theorem poses four such conditions, of which
a correct probabilistic description) would follow a prob- our model satisfies three, as does region I or II by itself.
ability distribution given by a WKB-type calculation of The fourth condition is that there exists a point P such
the decay rate. Near the transition time the system can- that there is a finite difference in volume between the
not be described classically; we must “shade out” the interior of the PLC of P and that of any point P ′ in
region where only a quantum description is accurate, as the past of P . This is motivated by double-well infla-
in Fig. 3. tion, and claimed as necessary for semi-eternal inflation
(the argument being that if this condition does not hold,
Bubble nucleation can be described semi-classically in then any inflating point will find itself in a bubble at
a similar way. Here, we must attach allowed classical so- the next instant, with probability one). We argue that
lutions of the field equations along some boundary that this condition is not quite necessary: what is required is
represents the nucleation event. To do this in a covari- that there be a finite volume in the region between the
ant way, this boundary must be a surface of zero proper PLCs of P and P ′ in which bubbles can nucleate toward
distance, i.e. a null cone, as shown in Fig 3. A bubble nu- P . Thus if, as in the proof of the theorem, the“past of P ”
cleation “event” is thus comprised of a region (shown as is taken to be the full volume interior to the light-cone
the shaded upper quadrant) where a classical bubble in- pointing in the time direction away from which a bubble
terior solution applies, attached to locally dS regions by nucleated at P would expand, then the fourth condition
a “shaded out” region (within some proper separation applies to region I alone (and correctly implies that it is
squared of the nucleation point) where only a quantum extendible). But it would not apply to the full spacetime
description is valid. To produce a semi-classical descrip- (which is neither extendible nor singular), because the
tion of a spacetime in which nucleation events occur, one relevant part of the light-cone extends only to J − .
must then populate it with these configurations in such a The argument of the newer singularity theorem [4] con-
way that the nucleation sites are randomly situated and sists of the definition of a local “Hubble parameter” H
occur at the correct rate per unit four-volume, while the meant to represent the rate of divergence of neighbor-
classical description, which applies far from the nucle- ing comoving test particles, along with an argument that
ation site, is in accord with the (classical) boundary con- any region in which a suitable average Hav of H is greater
ditions. When the classical boundary conditions are ours, than zero along all geodesics must be geodesically incom-
given on J − , this yields the bubble distribution indi- plete.
cated in Fig. 2. The (semi-classical) boundary conditions We understand Borde at al.’s argument as follows. One
do, then, control the time direction of bubble nucleation, imagines some timelike or null test geodesic with affine
not by introducing some locally-detectable AOT, but by parameter λ in the spacetime in question, then attempts
controlling the allowed global configuration of bubble nu- to construct a timelike vector field uµ (λ) along the test
cleation events. geodesic into its past such that Hav (defined via uµ ) ex-
ceeds zero. It is shown that this can only be achieved
along some finite affine length of the test geodesic, since
There is one final “region” in which we can check the imposed condition rapidly forces uµ towards nullness.
the AOT: the entire universe. Interestingly, we here Borde et al. [4] then take their result to mean that an
find that while each bubble nucleation event is non-time- eternally inflating spacetime is past-geodesically incom-
symmetric, by virtue of the symmetry of the cosmologi- plete.
cal boundary conditions, the statistical description of the We take the hypothetical satisfaction of their averaged
universe does admit a sort of T-symmetry. In Sec. V we Hubble parameter condition for all test geodesics as the
will discuss the possibility of making this symmetry ex- implicit definition of what Borde et al. mean by an eter-
act via an identification on the manifold. This raises the nally inflating spacetime. The logic is that uµ might be
intriguing possibility of having a well-defined (and con- independent of test geodesic and simply be the velocity
sistent among communicating observers) AOT for all ob- field of some set of comoving worldlines in the inflating
servers even while the physical laws and the global physi- spacetime. So what the theorem actually implies is that
cal description of the universe both admit a time-reversal is it impossible to entirely cover a spacetime with such a
symmetry. set of worldlines in a way that allows all test geodesics
7
cutting these worldlines to obey the Hubble parameter IV. BOUNDARY CONDITIONS AND THE
condition. NULL BOUNDARY PROPOSAL
For illustration, let us consider dS space. We note that
as pure dS is a maximally symmetric spacetime, it does We have seen how extending semi-eternal inflation to
not make sense to regard some parts as inflating and oth- eternal inflation implies particular behavior on the infi-
ers not—one must break the symmetry by adding some nite null surface J − . Here we discuss the converse, de-
other ingredients, and then frame a discussion in terms scribing how the eternal double-well inflation model we
of them. Nevertheless, let us investigate if the geodesics have described can be specified by a particularly simple
defined by having fixed comoving coordinates themselves set of cosmological boundary conditions that are imposed
constitute a uµ field satisfying Hav > 0. It turns out on J − .
that H reduces to the usual ȧ/a in this situation. In the
closed slicing with metric
A. On cosmological “initial” conditions
2
ds2 = −dt̃2 + H −2 cosh H t̃ dχ2 + sin2 χdΩ2 , 2
(7)
The correct specification even of a complete set of
physical laws does not by itself allow prediction of any
the coordinates cover all of dS, so uµ is globally defined. physical system’s behavior; these laws must be sup-
However, H < 0 for t̃ ≤ 0 so Hav goes negative there. plemented by boundary conditions that suffice to fully
Now consider using a single flat or open coordinate patch characterize the system being modeled. The Big-Bang
to define uµ . Hav > 0 here (at least for the appropri- (“BB”) model essentially consists of a set of such bound-
ate choice of time orientation). However, because nei- ary conditions for our observable cosmic surroundings:
ther of these coordinate patches covers the spacetime, at some early time, our region was a hot, dense, nearly
neither does uµ . Furthermore, one may choose an infi- homogeneous and isotropic mixture of particles and fields
nite number of flat coordinatizations of dS, each with a in thermal and chemical equilibrium, in a nearly-flat ex-
different null boundary where the uµ construction fails. panding background geometry with scale-invariant grav-
This makes it clear that the boundary of a given uµ field itational potential perturbations of amplitude 10−5 on
cannot be unambiguously used to define an edge to an the scale of the cosmological horizon.
inflating region. While these initial conditions yield predictions in ex-
We do not believe that the global existence of a suitable cellent accord with astronomical observations, they are
uµ is necessarily the best definition of what is meant by deemed by many to be too special: they seem to com-
an eternally inflating spacetime.4 In particular, equat- prise an extremely small portion of some “ensemble of
ing H with physical expansion entails a tacit assumption all possible initial conditions”, using some (generally un-
that the physical AOT is everywhere in accord with that specified) measure (see e.g. [21]). The theory of inflation
defined by uµ . The model we have proposed could be is widely accepted as a way to broaden the range of al-
covered by a congruence of geodesics (those comoving in low initial conditions by funneling a relatively wide class
two flat coordinate patches covering dS) that would yield of physical conditions into satisfactory BB initial condi-
H < 0 in some regions. However, we have argued that tions [22].
these regions may still be regarded as expanding with re- This is perhaps reasonable as a stop-gap measure as
spect to the physical AOT defined by the cosmological long as the true (pre-inflation) initial conditions are un-
boundary conditions. known. But it does not solve the initial condition prob-
In summary, both singularity theorems postulate con- lem, for not all initial conditions will give rise to infla-
ditions for a region to be “inflating”, and find that such tion [23], nor is it clear that all of those that do give
a region cannot be geodesically complete. However, in- inflation will yield a viable big-bang model [21]. Thus
terpreting these theorems as forbidding eternal inflation one must still assume that suitable conditions emerge
seem to us to require an unwarranted assumption about from the initial singularity; whether this assumption cor-
the global AOT independent of the cosmological bound- responds to a “special” or “generic” condition seems ill-
ary conditions. defined without a description of the singularity, which
would presumably require quantum gravity. But unless
quantum gravity completely alters the way physical the-
ories are applied, it is unlikely to yield a unique initial
4 Indeed the term “eternal inflation” has been used with a vari-
condition; it will still be necessary to supply a quantum
ety of meanings. For example, the recent paper [20] used it to state (or something similar). Even if it is possible to de-
describe models that are eternally inflating to the future, but fine the ensemble of all possible states, the question of
simply geodesically complete to the past (ȧ and/or ä may go how (and why) a state was “chosen” for our universe can
negative there). Such histories, with a globally-defined arrow of only be a metaphysical one, and it seems that, just as in
time, seem physically unrealistic with a typical mechanism for
exiting inflation, since this would render them unstable to the choosing physical laws, we can at best posit a state on
formation of thermalized regions in the putatively eternal early some (possibly symmetrical or aesthetic or philosophical)
phase[2]. grounds, derive its consequences, and compare them to
8
a semi-classical description (though we return to QFT in surfaces. For further details on both the WCH and NBP
Sec. V A). see [28].
The tunneling approach to quantum cosmology [29] ar-
gues that the semi-classical universe emerges via a quan-
D. Discussion of the null boundary proposal tum tunneling event. In the context of models with
open-inflationary potentials the proposal suggests that
The boundary conditions we have proposed are ex- the universe, when first semi-classically describable, is
tremely simple, in line with the view espoused above that most likely to be small and regular, with the field away
one cannot avoid making a specific choice for cosmolog- from the true vacuum; future-eternal inflation can then
ical boundary conditions, and that it is then reasonable ensue. The tunneling is supposed to have occurred out
to make a choice that is as simple and as highly symmet- of a quantum gravitational chaos so severe as to pre-
ric as possible, rather than hope to choose a “generic” clude any space-time description. While our proposal
boundary condition. also leads to inflation from semi-classical boundary con-
ditions, it explicitly avoids any such extreme quantum
gravitational regime.
1. Relation to other proposals Over the years Sakharov has discussed various cosmo-
logical models involving time-reversal (and CPT) invari-
ance [30, 31, 32, 33]. In these models, the universe is
Our proposal is novel in that it requires no treatment
generally symmetric across a singular FRW bounce at
of an initial singularity, and in that boundary conditions
t = 0, at which universe assumes an especially simple
are placed on a null boundary that does not correspond
state; away from this bounce entropy increases in both
to any particular cosmic “time”. It does, however, have
directions of time [31]. In [30] he hypothesizes that phe-
features in common with other proposals for specifying
nomena at t < 0 are the CPT reflections of the phenom-
the state of the universe.
ena at t > 0. In [31, 32] he considers the possibility of
Penrose’s Weyl Curvature Hypothesis (WCH) requires
an infinite chain of further bounces or cycles away from
that initial singularities are constrained to have vanishing
t = 0 in both time directions, and also the possibility
Weyl tensor, corresponding to a low-entropy state. This,
that the minimum-entropy surface might be one of max-
he argues, gives rise to an arrow of time flowing away
imum expansion rather than a singular one [31, 32]. In a
from the initial state. Our proposal is very similar, except
paper primarily about signature change [33] he alludes to
of course that we impose the vanishing of the Weyl tensor
possibility of non-singular time-reversal around a surface
on a lightlike slice running across the universe rather than
of minimal radius in a false vacuum state. Our proposal
at an initial singularity.
clearly has parallels with these ideas. We, however, con-
The Hartle-Hawking No Boundary Proposal (NBP) is
centrate on an infinite non-singular null surface, rather
a prescription for the wavefunction of the universe. It
than a singular spacelike one. Moreover, with the con-
provides Ψ(hij , φ), the amplitude for a given three-metric
cepts of semi-classical bubble nucleation and open infla-
and scalar field configuration on a spatial 3D surface.
tion, we are able to provide a relatively complete physical
The construction is designed to suppress irregular con-
picture.
figurations relative to more regular ones, and to thus fa-
vor simple and symmetric states. The notion of temporal
evolution does not appear explicitly. However, one can
2. The horizon problem
associate histories with saddle-point approximations to
the wavefunction, and it turns out that such universes
are often smooth when they are young and small. Thus Inflation was originally conceived as a remedy for
the NBP may be taken to imply that, at the semi-classical troubling issues concerning cosmological initial condi-
level, there should exist a spatial surface in a cosmological tions [34], thus it is useful to compare how such shortcom-
spacetime on which the boundary conditions are partic- ings of the HBB model [22] are dealt with in our model.
ularly simple.5 . Our proposal rather uses a null surface We focus on what is (aside from the singularity problem)
(but see Sec. VI B below). It would be very interesting to perhaps the most vexing of these HBB difficulties: the
develop a quantum prescription in a similar vein to the horizon problem.
NBP which naturally leads to simplicity on certain null The horizon problem is generally framed as fol-
lows: choose two spatially antipodal points on the last-
scattering surface. They are similar in temperature, yet
their PLCs never intersect in a HBB cosmology, so there
5 In the context of models with open-inflationary potentials, the can be no causal connection between them. Inflation is
NBP seems to favor histories in which the scalar field is every- generally thought to solve this problem, because with suf-
where in its true vacuum [26]. The NBP may still be relevant ficient inflation, the PLCs will intersect. But this does
for inflation with the use of an anthropic constraint, or in a
“top-down” approach to calculating quantum probabilities [27]. not suffice for the points to have the same temperature,
Starting off in the true vacuum is however not a problem for because the temperature at each point depends on data
recycling models of inflation, as discussed below. across its full PLC, and there is a portion of each PLC
10
that does not intersect the other. Taking this into ac- problem (posed in terms of Hubble volumes within the
count, for the two points to have similar properties it boundary condition surface) to approximately the same
must be assumed that there is sufficient inflation, and degree as does inflation beginning in the Planck epoch,
additionally that at inflation’s beginning at time tinf , the and much better than does inflation with an early quasi-
region to the past of the two points is homogeneous on FRW phase.
−1
length scales of order Hinf . But suppose there is a earlier
epoch between some time t and tinf . Then at t the patch
must be homogeneous over a physical length scale of at V. THE ELLIPTIC VIEW
least
a(t)H(t) The model we have proposed consists of two indistin-
r(t) ∼ H −1 (t), (8) guishable regions, each comprising an eternally inflating
a(tinf )H(tinf ) universe with an AOT (where defined) pointing away
where H(t) ≡ ȧ/a. If a ∝ tα (α = 1/2 for radiation- from an infinite null surface which connects the two re-
dominated expansion) prior to tinf , then r/(H −1 ) ≃ gions. The statistical identity of these universes, along
[a/a(tinf )](α−1) . Thus it is necessary to postulate that with lack of a global physical time orientation, suggests
our region was, at some initial time t0 , homogeneous over some form of an old idea concerning dS, called “the ellip-
[a(tinf )/a(t0 )](3−3α) Hubble volumes. The horizon prob- tic interpretation”8 that would identify the two universes.
lem therefore persists if tinf > t0 and α < 1. There are The idea consists of deeming an event to be rep-
two escapes available within inflation. The first is to set resented not by a single point of a spacetime mani-
tinf = t0 = tpl , so that the expansion is inflationary all fold, but by a pair of antipodal points (defined below).
the way back to the Planck time, before which one can- This corresponds to a topological identification that,
not speak of the expansion at all. The horizon problem applied to our model, identifies regions I and II, and
is then greatly ameliorated, as it must only be assumed maps J − onto itself (the R × S 2 manifold J − becomes
that some regions “emerging” from the Planck epoch are (R × S 2 )/Z2 ). This identification has been subject of
homogeneous over a relatively small (but > 1; see [23]) some previous [36, 37, 38, 39] and recent [40] investiga-
number of Hubble volumes when they are first classically tions.
describable; but whether this constitutes a true solution Pure dS can be represented as a 4D hyperboloid
will be seen only if and when the Planck epoch itself is
−v 2 + w2 + x2 + y 2 + z 2 = H −2 (9)
understood. A second potential escape is to set α ≥ 1, or
(more generally) for ȧ to be non-decreasing, i.e. to have embedded in 4+1 D Minkowski space with metric
past-eternal inflation.
How, then, does the horizon problem look in the con- ds2 = −dv 2 + dw2 + dx2 + dy 2 + dz 2 . (10)
text of our eternal model with no initial time? Let us pose
the problem in a slightly more general manner: when The elliptic interpretation consists of identifying each
specifying the cosmological boundary conditions, must point P with coordinates (v, w, x, y, z) with its antipode
one do so over a region that is very large compared to −P at (−v, −w, −x, −y, −z). In the conformal diagram,
some relevant physical volume such as H −3 ? Posed this this means that points such as P and −P of Fig. 2 are
way, it might seem that because J − is an infinite sur- physically identified; the antipodal map A looks like a
face, a strong horizon problem exists.6 This, however, is vertical reflection and a horizontal shift through one half
not necessarily the case. Imagine J − as the limit of a se- of the diagram’s horizontal extent (Note that although
quence of spacelike slices obtained by boosting the space- this map makes an orbifold of the embedding space, there
like surface given by t̃ = 0 in global coordinates. Because
each such surface has volume 2π 2 H −3 , we might also at-
tribute to J − the same finite invariant volume.7 Thus
our construction would seem to ameliorate the horizon −v1 v2 +w1 w2 +x1 x2 +y1 y2 +z1 z2 [11]. In general −∞ < D < ∞,
whereas for two points on J − , |D| ≤ 1; they may thus be con-
sidered “close”. The relation between D and geodesic distance is,
unfortunately, not always defined. The points are timelike or null
separated, respectively, for D > 1 or D = 1. For |D| < 1 points
6 This is apparently the case, for example, in the cyclic model [14], are connectable by a spacelike geodesic of length cos−1 (D) < π;
where if the cyclic behavior is to continue indefinitely into the but for D ≤ −1 there is no geodesic connecting P1 and P2 , al-
future and past, it seems necessary to place cosmological bound- though one does connect P1 to the antipode P¯2 of P2 because
ary confitions on an infinite spacelike surface. This specification D(P1 , P¯2 ) = −D(P1 , P2 ). Under the identification of antipodal
accords the same properties to points that are arbitrarily distant, points (see below), any P1 and P2 are null-separated or spacelike-
and causally disconnected. separated by a geodesic distance < π/2H.
7 One can also consider other ways of taking a limit to J − , but 8 The etymology of this term is obscure to the authors. The el-
this one seems most in accord with the symmetries of dS. An- liptic view applied to spatial sections of dS was discussed first
other approach to the problem is to rather consider some dis- by de Sitter, who preferred it to the (now) conventional “spher-
tance measure between any two points on J − . A dS-invariant ical” view; and cited a letter from Einstein voicing the same
quantity, in units where H = 1, between points P1 and P2 in preference [35]. Antipodally identifying in space and time was
the embedding space (see Eq.(9) below) is given by D(P1 , P2 ) = discussed in fond detail by Schrödinger [36].
11
made to have the correct symmetry. The problem arises on J − . Since the lowest vacuum state must be posi-
when both points are outside of the causal diamond and tive for this scenario to work, it also has the potential
its copy; in this case the commutator turns out to van- to connect with the presently observed positive vacuum
ish for timelike separated points, and not for spacelike energy.
separated points. It is unclear whether this makes sense. A somewhat similar scenario can be realized using
In short, defining a satisfactory QFT in edS is rather “chaotic inflation” potentials such as V (φ) = λφ4 + V0 .
difficult; the difficulties stem primarily from the commu- Here we set the inflaton to rest at φ = 0 on J − , and
tator function, because it is not symmetric under time- require V0 > 0 so that the space is locally dS near J − .
reflection, while the anti-commutator is. It is then diffi- This is, again, a stable configuration, yet regions of large
cult for both functions to be symmetric under A, which potential may “nucleate” in this dS background as highly
includes time-reflection. It is possible [40] that string improbable quantum fluctuations in the field. If such a
theory in edS will make more sense than in dS. It is also region nucleates at high enough φ, it would provide the
conceivable that the elliptic view could emerge from a potential seed for the sort of eternal inflation envisioned
correct quantum treatment of dS, and that the described by Ref. [46], in which quantum fluctuations in φ domi-
troubles stem from doing QFT on a fixed background not nate over classical rolling so that inflation become eter-
included in the dynamics. This would be an interesting nal. While the structure of the resulting region becomes
issue to pursue in string theory or other theories of quan- extremely complicated, it can again be contained within
tum gravity. For now we must leave it there, and return a light-cone of some point in the original (dS) embedding
to eternal inflation. space, so that the global structure of the universe (filled
with such light-cones) is still understandable, and again
looks like Fig. 2.
VI. GENERALIZATIONS AND EXTENSIONS
VII. SUMMARY AND CONCLUSIONS on the manifold. We, and others, have studied this iden-
tification on de Sitter space classically and quantum-
We have investigated the possibility of making “future- mechanically. Antipodally identified (or “elliptic”) dS
eternal” inflation eternal also to the past. Starting with has the virtues that it is causally stable and observers
a de Sitter spacetime background dominated by the false- have no event horizons. It has the disadvantage that its
vacuum energy in which true-vacuum bubbles can form, non-time-orientability makes defining a reasonable quan-
we specify that the bubble distribution at any time is tum field theory difficult. With antipodal identification
in exactly the steady-state configuration asymptotically our model is more economical as the two duplicate uni-
approached by semi-eternal inflation. All bubbles are verses are identified; however not all of the attractive
then equivalent, and the statistical distribution of bub- features of “pure” edS remain.
bles admits both a time-translation and space-translation Our model can be generalized to other inflaton po-
invariance in the background inflating space. tentials (such as for chaotic inflation), and therefore al-
The described region (“region I”) has no initial time, lows one to partially understand the global structure of
but has a null boundary J − that is the limiting surface the eternally inflating spaces that result. One may also
as t → −∞ of the flat spatial sections. The steady- use an analogous construction to specify a semi-eternally
state configuration at t > −∞ corresponds to the infla- inflating but non-singular universe by placing boundary
ton field being in the false vacuum state everywhere on conditions on a spacelike section.
J − , with no incoming radiation from J − . This sur- Our primary conclusion is that it is possible, using only
face can be reached by a past-directed geodesic of finite “standard” ingredients underlying popular models for in-
proper length from any point in the space, so region I is flation, to specify simple cosmological boundary condi-
geodesically incomplete; this is the “singularity” pointed tions on an infinite null surface that lead to past- and
to by theorems purporting that past-eternal inflation is future-eternal inflation. Such a universe would obey the
impossible [3, 4]. But the space can (and should) be ex- same Perfect Cosmological Principle that governs a semi-
tended, as the state on J − also constitutes boundary con- eternally inflating universe long after its beginning. If our
ditions for the region past J − if the manifold is extended. construction survives scrutiny, and can be specified at the
These boundary conditions imply that a duplicate copy fully quantum (or quantum-gravitational) level within a
of region I exists on the other side of J − . Together re- theory of fundamental physics, it could serve as the basis
gion I and the new “region II” constitute a geodesically for a realistic cosmology that avoids a cosmological sin-
complete (i.e. non-singular) inflating cosmology that ad- gularity, a beginning of time, or a creation of the universe
mits a coordinatization in which the background space “from nothing”.
and bubble distribution are time-translation and space- Acknowledgments
translation invariant.
The null surface J − constitutes a cosmological bound- We thank Alex Vilenkin for helpful comments on a
ary condition surface on which the universe is classically draft of this paper. AA is supported by a grant from the
in a particularly simple and symmetric state. Although W.M. Keck foundation. SG is supported in part by US
J − is an infinite null surface, some of its points are in Department of Energy grant DE-FG02-91ER40671.
causal contact, and one might attribute a finite invariant
volume of O(H −3 ) to it. Therefore specifying boundary
APPENDIX
conditions on it is not like specifying them on an infi-
Here we review Green functions and the initial value
nite spacelike hypersurface (which would lead to a severe
problem in some detail, focusing for simplicity on the
“horizon problem”).
inhomogeneous scalar wave equation in fixed background
It might be possible to construct a quantum state cor- spacetimes. We are interested in how much information
responding to our classical boundary conditions on J − by must be specified, and where, in order to fix the field
taking a null-limit of spacelike sections on which the wave throughout spacetime.
functional describing the fields is centered on the desired
classical state. But an explicitly null quantum formula-
tion of our null boundary proposal has not been provided 1. Defining Green functions
and would constitute an interesting future study.
It is widely thought that the “arrow of time” is con-
Imagine that our scalar field φ satisfies
nected with cosmological boundary conditions. In our
model, time flows away from J − , and the AOT is con- − m2 φ = q (xµ ) ,
(11)
sistent among all physical observers that can compare it.
The AOT is not, however, defined globally, and in our where q is an arbitrary source term, and we wish to de-
model the statistical description of the universe admits a termine φ at some point P with coordinates xP . This is
global symmetry that includes time-reversal. possible given φ and its time derivative on some complete
This symmetry, along with the presence of two dupli- spacelike slice through the past light-cone (PLC) of P .
cate, non-communicating universes, motivates—though Let us see why this is so, and whether this is the only
does not require—formally identifying antipodal points suitable set of initial data for the problem.
15
with coordinates such that P is at the origin. That coordinate patch of the open slicing of dS. The equation
this solves φ = 0 can be seen by expanding out the reads:
δ-function and comparing to a general superposition of
1
incoming and outgoing spherically symmetric waves. (A ∂t sinh3 Ht∂t φ + m2 φ = 0.
3 (17)
nice discussion of this function is found in [48].) Note sinh Ht
that this is only non-zero on the light-cone itself, vanish-
To solve this [50], write φ = Hχ/ sinh Ht and set z =
ing even inside the cone. This is a special property of
cosh Ht to obtain
massless fields in even-dimensional spacetimes (see [15]
for more details.) m2
1
To obtain the solution for the massive field in (1 − z 2 )χ,zz − 2zχ,z + 2 − 2 − χ = 0, (18)
H 1 − z2
Minkowski space, we start by considering
which is Legendre’s equation [49] with ν = −1/2 +
G1 ≡ f t2 − r2 sgn (t) Θ t2 − r2 .
p
9/4 − m2 /H 2 and µ = −1. The solution to this equa-
tion which is regular as z → 1 is Pνµ (z). In terms of t,
The sgn and Θ functions enforce the gross features of this tends to Ht/2 as t → 0 independent of m2 , hence φ
the behaviour that we require. Writing using spherical tends to H/2.
polar coordinates, we find We can now use this result to deduce the form of the
commutator over all of dS space. Extend the meaning of
( − m2 )G1 = sgn (t) Θ t2 − r2 − m2 f
τ 2 to be the signed geodesic distance squared between P
− 4f (0) sgn(t)δ t2 − r2 .
(15) and the point in question. Note that we do not consider
the point to be in the light-cone of the antipode of P ,
Let us choose f (τ 2 ) (where here τ 2 = t2 − r2 ) such that since no geodesic exists which connects it to P . In this
2
− m f = 0 inside the light-cones; a general solution case we rather define GC to be zero (but see our discus-
is f (τ 2 ) = AJ1 (mτ )/τ + BY1 (mτ )/τ where J1 and Y1 are sion of the antipodal quantum commutator above.) Near
Bessel functions as in Ref. [49]. The first term of Eq. (15) P the space is locally Minkowskian, so we can compare
is then zero. For the second term, we notice that it is to our Minkowksi result to obtain:
basically the solution to the massless problem that we’ve
already found. So let us simply add the massless solution
m2 Θ(H 2 τ 2 ) −1
GM M0 sgn(t)
C from Eqn. (14) to our ansatz G1 . Since GC = 0, δ(τ 2 ) −
0
GdS
C = Pν (cosh Hτ ) .
2 M0
we have left −m GC on the RHS. By choosing A = 2π 2 sinh Hτ
−m/4π and B = 0, we obtain a complete cancellation, (19)
leaving us with our desired solution. Thus the full Green Here by t is meant some suitable generalization of
function is Minkowski time, with sgn(t) thus serving to make GdS C
be of opposite
√ sign in the future and past light-cones, and
mΘ(τ 2 )J1 (mτ ) τ = + τ 2 . The sgn(t)δ(τ 2 ) should be interpreted as the
sgn (t) 2
GMC = δ τ − . (16) generalization of the equivalent term in the Minkowski
2π 2τ
result.
We may now consider generalizing Eq. (16) to other
spacetimes such as dS. First, let us note that we took
f above to be a function of the proper time τ from the 3. Domains of dependence
origin, within the light-cone. Thus the commutator func-
tion was invariant under Lorentz boosts. This suggests Given our Green functions, we would like to investigate
that in dS, for example, we make our commutator inside to what extent fixing the fields on a null-cone L of a
the forward light-cone invariant under isometries which point Q determines the field values within that cone. We
leave P fixed. This is the same as saying that it should be are particularly interested in the importance of the field
a function of the proper time from the origin alone. One near Q, as our cosmological boundary surface J − can
can define this quantity in terms of an “angle” in the 5D be considered the light-cone with Q “at” past-timelike
embedding space [11] (see footnote to Sec IV D 2), but infinity.
perhaps a more intuitive way to proceed is as follows. A Take, as a first example, the Green function GR for a
patch of dS can be covered by coordinates in which the massless field such as the scalar field in 4D Minkowski
metric is the same as that for an open expanding FRW space with GR from Eq. (14). Because GR has support
universe, with scale factor H −1 sinh Ht. The proper time only on the PLC of P , the field at P depends only on the
along the geodesics representing comoving observers is intersection of P ’s PLC and L. This indicates two things.
just given by the coordinate time t. These geodesics all First, specifying data on L explicitly determines the field
intersect at a point as t → 0, which we choose to be everywhere inside L (i.e. everywhere in the future of Q).
P , and they cover all of the interior of the FLC of P . Second, the field at any point P inside L does not de-
We therefore need only find an appropriate spatially ho- pend on φ(Q); we might thus consider Q to be irrelevant
mogeneous solution of the massive wave equation in the in terms of what occurs within L. This, however, hides
17
a subtlety: while Gr and φ on L always allow the con- the limit τ → ∞). More explicitly, we may consider a
struction of a valid solution of the field equations, there boundary surface comprised of J − at t̃ > t̃0 for some
is no guarantee that the field so obtained is continuous t̃0 , closed by the spacelike surface t̃ = t̃0 , where t̃ and
with the field specified on L unless some assumptions of t̃0 are in the global closed coordinates. Then the field
field regularity at the vertex (Q) are made: if we wish integral (13) consists of an integral along J − , and an
to evaluate φ on L to check that we have a genuine solu- integral over the t̃ = t̃0 hypersurface, which has a finite
tion to our boundary value problem, then we must either physical volume of order H −3 . But then as t̃0 → −∞,
know φ(Q), or assume that the field is regular as Q is τ → ∞, so this second contribution will vanish unless
approached along L, so that we can extend the field to either the average of the field or its t̃-derivative blow up
−1
Q by continuity. faster than GR . Hence we expect that the field at any
For a massive field the strict “Huygen’s Principle” does point in the space will depend only on the values on J − ,
not hold; while much of the contribution to the integral and not on the field behavior at past infinity, as long as
in Eq.(13) comes from the PLC of P where GR is singular the fields are assumed to be suitably finite and regular.
(see [15] for some discussion of this “generalized Huygen’s This argument might also be applicable to gravity; the
principle”), there is also a contribution from inside P ’s equations governing the Weyl conformal tensor can be
PLC because GR is everywhere nonzero there. cast, at the linear level, as wave-equations for a spin-2
In the dS case, GR as given by Eq. (19) falls off expo- field, and boundary conditions can be specified on a null
nentially as τ → ∞; thus we may expect that when L is surface such as J − [24]. We therefore expect that our
taken to be J − , where the vertex lies “at” τ = ∞, the boundary conditions on J − determine the Weyl tensor
field values inside J − will not depend on the field at the uniquely, and therefore determine the spacetime to be
vertex (where by “at the vertex” we mean within J − in pure dS when bubble nucleations are neglected.
[1] A. Linde, D. Linde and A. Mezhlumian, Phys. Rev. D [18] See, e.g., H.D. Zeh, “The physical basis of the direction
54, 2504 (1996); A. Vilenkin, Phys. Rev. D 46, 2355 of time”, (Berlin: Springer-Verlag, 1999).
(1992). [19] R. Penrose, in “Quantum Gravity 2: A Second Oxford
[2] See, for example, A.H. Guth, Phys. Rept. 333, 555 (2000) Symposium”, ed. C.J. Isham, et al. (Oxford: Clarenden
and references therein; A.D. Linde, Phys. Lett. B175, Press), 1981.
395 (1986); A.D. Linde in The Very Early Universe, [20] G.F.R. Ellis and R. Maartens, qr-qc/0211082.
edited by G.W. Gibbons, S.W. Hawking and S.T.C. Sik- [21] R. Penrose, in “Fourteenth Texas Symposium on Rel-
los (Cambridge University Press, Cambridge, England, ativistic Astrophysics, Dallas, Tex., 1988” (New York:
1983); P.J. Steinhardt, same volume. Annals of the New York Academy of Sciences, 1989); S.
[3] For a review, see A. Borde and A. Vilenkin, Int. J. Mod. Hollands and R. Wald; gr-qc/0205058; N. Turok, Class.
Phys. D5 (1996) 813-824. Quantum Grav. 19, 3449 (2002).
[4] A. Borde, A.H. Guth and A. Vilenkin, gr-qc/0110012. [22] A. Linde, “Particle physics and inflationary cosmology”,
[5] A. Aguirre and S. Gratton, Phys. Rev. D 65 (2002) (Switzerland: Harwood Acedemic Publishers, 1990).
083507. [23] T. Vachaspati and M. Trodden, Phys. Rev. D 61, 023502
[6] A.H. Guth, Phys. Rev. D 23, 347 (1981). (2000).
[7] M. Bucher, A.S. Goldhaber and N. Turok, Phys. Rev. D [24] R. Penrose and W. Rindler, “Spinors and Space-Time,
52, 3314 (1995). Vol. I”, (Cambridge: Cambridge University Press, 1984).
[8] J.R. Gott, Nature (London) 295, 304 (1982); K. Ya- [25] T. Heinzl, Lect. Notes Phys. 572, 55 (2001),
mamoto, M. Sasaki, and T. Tanaka, Astrophys. J. 455, hep-th/0008096.
412 (1995); A. Linde and A. Mezhlumian, Phys. Rev. D [26] S. Gratton and N. Turok, Phys. Rev. D 63, 123514
52, 6789 (1995) (2001).
[9] A. Vilenkin, Phys. Rev. D 46, 2355 (1992). [27] S.W. Hawking and T. Hertog, Phys. Rev. D 66, 123509
[10] A.H. Guth and E.J. Weinberg, Nucl. Phys. B212, 321 (2002).
(1983). [28] S.W. Hawking and R. Penrose, The Nature of Space and
[11] For reviews of de Sitter space, see: Ref. [41]; M. Spradlin, Time (Princeton University Press, Princeton, New Jer-
A. Strominger, and A. Volovich, hep-th/0110007. sey, USA, 1996).
[12] S. Coleman and F. De Luccia, Phys. Rev. D 21, 3305 [29] A. Vilenkin, Phys. Rev. D 27, 2848 (1983).
(1980). [30] A. D. Sakharov, JETP Lett. 5, 24 (1967).
[13] B. Mandelbrot, The Fractal Geometry of Nature, (W.H. [31] A. D. Sakharov, Sov. Phys. JETP 52, 349 (1980).
Freeman and Co., New York, 1983). [32] A. D. Sakharov, Sov. Phys. JETP 56, 705 (1982).
[14] P.J. Steinhardt and N. Turok, hep-th/0111030. [33] A. D. Sakharov, Sov. Phys. JETP 60, 214 (1984).
[15] R. Courant, Methods of Mathematical Physics: Vol. II, [34] A. H. Guth, Phys. Rev. D 23, 347 (1981).
Partial Differential Equations (New York: Interscience [35] W. de Sitter, M.N.R.A.S 78,3 (1917).
Publishers, John Wiley & Sons, 1962). [36] E. Schrödinger, Expanding Universes (Cambridge Uni-
[16] R. Penrose, Gen. Rel. and Grav. 12, 225 (1980). versity Press, Cambridge, England, 1956).
[17] V.A.Rubakov and S.M.Sibiryakov Theor. Math. Phys. [37] H. Schmidt, Fortschr. Phys. 41, 179 (1993).
120, 1194 (1999). [38] G. Gibbons, Nuc. Phys. B271, 497 (1986).
18
[39] N. Sanchez, Nuc. Phys. B294, 1111 (1987); J.L. Fried- (1998).
man and A. Higuchi, Phys. Rev. D 52, 5687 (1995). [46] A. Linde, D. Linde and A. Mezhlumian, Phys. Rev. D
[40] M. Parikh, I. Savonije and E. Verlinde, hep-th/0209120. 49, 1783 (1994).
[41] S.W. Hawking and G.F.R. Ellis, The large scale structure [47] R. Bousso, Rev. Mod. Phys. 74, 825 (2002).
of space-time (Cambridge University Press, Cambridge, [48] P.A.M. Dirac, Lectures on Quantum Field Theory, (New
England, 1973). York: Belfer Graduate School of Science), 1996.
[42] F. Leblond, D. Marolf and R. C. Myers, JHEP 0206, 052 [49] M. Abramowitz and I. Stegun, “Handbook of Mathemat-
(2002). ical Functions” (New York: Dover Publications, 1972).
[43] E. Mottola, Phys. Rev. D 31, 754 (1985); B. Allen, Phys. [50] M. Sasaki, T. Tanaka and K. Yamamoto, Phys. Rev. D
Rev. D 32, 3136 (1985). 51, 2979 (1995).
[44] K. Lee and E.J. Weinberg, Phys. Rev. D 36, 1088 (1987).
[45] J. Garriga and A. Vilenkin 1998, Phys. Rev. D 57, 2230